Faith in parsimony
mesibov at SOUTHCOM.COM.AU
Sat Jul 23 09:37:21 CDT 2005
Please allow for some stupidity here, as I'm new to this and my statistics
are the sort used outside computational taxonomy.
However, as I understand it, bootstrapping cannot logically give
posterior-based confidence limits for a tree derived from a particular data
matrix, because it looks at trees derived from random resamplings of that
matrix: pseudoreplicate matrices.
Bremer support, as I understand it, is based on looking for shortest trees
given certain topological constraints (a set of constraint trees). This
again does not sample the possible universe of trees based on a particular
I guess my puzzlement centers around this: there are two kinds of
uncertainty with parsimony-based trees. The "easier" uncertainty concerns
whether or not the shortest tree or trees derived from a matrix is (are)
indeed the shortest tree(s). If you haven't done an exhaustive search, you
don't know, and you need a measure of your uncertainty.
The "harder" uncertainty concerns whether or not your shortest tree(s) is a
reasonable guess at what actually happened in evolution. It seems to me
perfectly logical to posit a model for that evolution, generate a large set
of trees based on the model, and see where the shortest tree(s) fit(s) in
the resulting frequency distribution. This is how posterior probabilities
and confidence limits are estimated in other areas of statistics. Am I
Richard Zander wrote:
"Both maximum parsimony and Bayesian analysis can be used for morphological
Can you point to a Bayesian example? I'm particularly interested in what
sorts of prior modeling of morphological evolution were done before a Markov
chain Monte Carlo simulation was attempted.
I have also found a paper:
Alfaro, M.E., Zoller, S. and Lutzoni, F. 2003. Bayes or Bootstrap? A
Simulation Study Comparing the Performance of Bayesian Markov Chain Monte
Carlo Sampling and Bootstrapping in Assessing Phylogenetic Confidence.
Molecular Biology and Evolution 20(2): 255-266.
which compares Bayesian-McMC, maximum likelihood bootstrap and maximum
parsimony bootstrap for assessing "confidence", and finds practical
advantages to the Bayesian approach. There is also a good introduction in
this paper to what I now realise is a busy area of research.
Dr Robert Mesibov
Honorary Research Associate, Queen Victoria Museum and Art Gallery
and School of Zoology, University of Tasmania
Home contact: PO Box 101, Penguin, Tasmania, Australia 7316
(03) 6437 1195
Spatial data basics for Tasmania
More information about the Taxacom